Observability

Built-in metrics and logs, plus how to forward telemetry to your existing stack.

Build logs stream on the Logs tab through Deploy live and Healthy, then the Metrics tab shows CPU and memory tiles, requests per minute, 5xx rate, and p95 latency.

Coroot

Observability is Coroot (eBPF metrics, logs, and traces). Open it from an app in the dashboard — StackBlaze mints a one-time handoff and opens Coroot in a new tab. Billing and in-dashboard charts query PromQL query_range with an API key; there is no instant /query endpoint.

Signal	Use it for
CPU and RSS memory	Right-sizing pod sizes and catching OOMs
TCP egress (eBPF)	Unexpected outbound traffic
Golden signals / incidents	Error rate and latency regressions after a deploy
Service map	Who talks to whom inside the phase

Logs

Runtime stdout/stderr is in the app Logs tab and via GET /api/logs/{pipeline}/{phase}/{app}/. Search uses the durable log index (search_logs in MCP / chat). An empty live tail is not the same as "no logs" — search history when the pod has rotated.

Note

Chat and MCP should prefer log search over a raw tail. Retention follows the observability backend for the region, not a per-plan hobby/pro table.

Notifications

Configure Slack, Discord, or generic webhooks under Settings → Notifications. Pick pipelines and event names (deploys, failures). This is not a full PagerDuty product — use Coroot incidents for SLO-style alerts.

Your own exporters

You can still emit OpenTelemetry from the app to Datadog, Honeycomb, or Grafana Cloud. That is in-process and independent of Coroot.

After a bad deploy

Open Coroot from the app, then roll back from Rollbacks or POST /api/deployments/rollback/{pipeline}/{phase}/{app}/{revisionId}.

Deep dive: Observability beyond metrics